[JFFS2] Reduce time for which erase_free_sem is held during erase.
[safe/jmp/linux-2.6] / fs / jffs2 / erase.c
1 /*
2  * JFFS2 -- Journalling Flash File System, Version 2.
3  *
4  * Copyright © 2001-2007 Red Hat, Inc.
5  *
6  * Created by David Woodhouse <dwmw2@infradead.org>
7  *
8  * For licensing information, see the file 'LICENCE' in this directory.
9  *
10  */
11
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/mtd/mtd.h>
15 #include <linux/compiler.h>
16 #include <linux/crc32.h>
17 #include <linux/sched.h>
18 #include <linux/pagemap.h>
19 #include "nodelist.h"
20
21 struct erase_priv_struct {
22         struct jffs2_eraseblock *jeb;
23         struct jffs2_sb_info *c;
24 };
25
26 #ifndef __ECOS
27 static void jffs2_erase_callback(struct erase_info *);
28 #endif
29 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset);
30 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
31 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
32
33 static void jffs2_erase_block(struct jffs2_sb_info *c,
34                               struct jffs2_eraseblock *jeb)
35 {
36         int ret;
37         uint32_t bad_offset;
38 #ifdef __ECOS
39        ret = jffs2_flash_erase(c, jeb);
40        if (!ret) {
41                jffs2_erase_succeeded(c, jeb);
42                return;
43        }
44        bad_offset = jeb->offset;
45 #else /* Linux */
46         struct erase_info *instr;
47
48         D1(printk(KERN_DEBUG "jffs2_erase_block(): erase block %#08x (range %#08x-%#08x)\n",
49                                 jeb->offset, jeb->offset, jeb->offset + c->sector_size));
50         instr = kmalloc(sizeof(struct erase_info) + sizeof(struct erase_priv_struct), GFP_KERNEL);
51         if (!instr) {
52                 printk(KERN_WARNING "kmalloc for struct erase_info in jffs2_erase_block failed. Refiling block for later\n");
53                 down(&c->erase_free_sem);
54                 spin_lock(&c->erase_completion_lock);
55                 list_move(&jeb->list, &c->erase_pending_list);
56                 c->erasing_size -= c->sector_size;
57                 c->dirty_size += c->sector_size;
58                 jeb->dirty_size = c->sector_size;
59                 spin_unlock(&c->erase_completion_lock);
60                 up(&c->erase_free_sem);
61                 return;
62         }
63
64         memset(instr, 0, sizeof(*instr));
65
66         instr->mtd = c->mtd;
67         instr->addr = jeb->offset;
68         instr->len = c->sector_size;
69         instr->callback = jffs2_erase_callback;
70         instr->priv = (unsigned long)(&instr[1]);
71         instr->fail_addr = 0xffffffff;
72
73         ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
74         ((struct erase_priv_struct *)instr->priv)->c = c;
75
76         ret = c->mtd->erase(c->mtd, instr);
77         if (!ret)
78                 return;
79
80         bad_offset = instr->fail_addr;
81         kfree(instr);
82 #endif /* __ECOS */
83
84         if (ret == -ENOMEM || ret == -EAGAIN) {
85                 /* Erase failed immediately. Refile it on the list */
86                 D1(printk(KERN_DEBUG "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n", jeb->offset, ret));
87                 down(&c->erase_free_sem);
88                 spin_lock(&c->erase_completion_lock);
89                 list_move(&jeb->list, &c->erase_pending_list);
90                 c->erasing_size -= c->sector_size;
91                 c->dirty_size += c->sector_size;
92                 jeb->dirty_size = c->sector_size;
93                 spin_unlock(&c->erase_completion_lock);
94                 up(&c->erase_free_sem);
95                 return;
96         }
97
98         if (ret == -EROFS)
99                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: -EROFS. Is the sector locked?\n", jeb->offset);
100         else
101                 printk(KERN_WARNING "Erase at 0x%08x failed immediately: errno %d\n", jeb->offset, ret);
102
103         jffs2_erase_failed(c, jeb, bad_offset);
104 }
105
106 void jffs2_erase_pending_blocks(struct jffs2_sb_info *c, int count)
107 {
108         struct jffs2_eraseblock *jeb;
109
110         down(&c->erase_free_sem);
111
112         spin_lock(&c->erase_completion_lock);
113
114         while (!list_empty(&c->erase_complete_list) ||
115                !list_empty(&c->erase_pending_list)) {
116
117                 if (!list_empty(&c->erase_complete_list)) {
118                         jeb = list_entry(c->erase_complete_list.next, struct jffs2_eraseblock, list);
119                         list_del(&jeb->list);
120                         spin_unlock(&c->erase_completion_lock);
121                         up(&c->erase_free_sem);
122                         jffs2_mark_erased_block(c, jeb);
123
124                         if (!--count) {
125                                 D1(printk(KERN_DEBUG "Count reached. jffs2_erase_pending_blocks leaving\n"));
126                                 goto done;
127                         }
128
129                 } else if (!list_empty(&c->erase_pending_list)) {
130                         jeb = list_entry(c->erase_pending_list.next, struct jffs2_eraseblock, list);
131                         D1(printk(KERN_DEBUG "Starting erase of pending block 0x%08x\n", jeb->offset));
132                         list_del(&jeb->list);
133                         c->erasing_size += c->sector_size;
134                         c->wasted_size -= jeb->wasted_size;
135                         c->free_size -= jeb->free_size;
136                         c->used_size -= jeb->used_size;
137                         c->dirty_size -= jeb->dirty_size;
138                         jeb->wasted_size = jeb->used_size = jeb->dirty_size = jeb->free_size = 0;
139                         jffs2_free_jeb_node_refs(c, jeb);
140                         list_add(&jeb->list, &c->erasing_list);
141                         spin_unlock(&c->erase_completion_lock);
142                         up(&c->erase_free_sem);
143
144                         jffs2_erase_block(c, jeb);
145
146                 } else {
147                         BUG();
148                 }
149
150                 /* Be nice */
151                 yield();
152                 down(&c->erase_free_sem);
153                 spin_lock(&c->erase_completion_lock);
154         }
155
156         spin_unlock(&c->erase_completion_lock);
157         up(&c->erase_free_sem);
158  done:
159         D1(printk(KERN_DEBUG "jffs2_erase_pending_blocks completed\n"));
160 }
161
162 static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
163 {
164         D1(printk(KERN_DEBUG "Erase completed successfully at 0x%08x\n", jeb->offset));
165         down(&c->erase_free_sem);
166         spin_lock(&c->erase_completion_lock);
167         list_move_tail(&jeb->list, &c->erase_complete_list);
168         spin_unlock(&c->erase_completion_lock);
169         up(&c->erase_free_sem);
170         /* Ensure that kupdated calls us again to mark them clean */
171         jffs2_erase_pending_trigger(c);
172 }
173
174 static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
175 {
176         /* For NAND, if the failure did not occur at the device level for a
177            specific physical page, don't bother updating the bad block table. */
178         if (jffs2_cleanmarker_oob(c) && (bad_offset != 0xffffffff)) {
179                 /* We had a device-level failure to erase.  Let's see if we've
180                    failed too many times. */
181                 if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
182                         /* We'd like to give this block another try. */
183                         down(&c->erase_free_sem);
184                         spin_lock(&c->erase_completion_lock);
185                         list_move(&jeb->list, &c->erase_pending_list);
186                         c->erasing_size -= c->sector_size;
187                         c->dirty_size += c->sector_size;
188                         jeb->dirty_size = c->sector_size;
189                         spin_unlock(&c->erase_completion_lock);
190                         up(&c->erase_free_sem);
191                         return;
192                 }
193         }
194
195         down(&c->erase_free_sem);
196         spin_lock(&c->erase_completion_lock);
197         c->erasing_size -= c->sector_size;
198         c->bad_size += c->sector_size;
199         list_move(&jeb->list, &c->bad_list);
200         c->nr_erasing_blocks--;
201         spin_unlock(&c->erase_completion_lock);
202         up(&c->erase_free_sem);
203         wake_up(&c->erase_wait);
204 }
205
206 #ifndef __ECOS
207 static void jffs2_erase_callback(struct erase_info *instr)
208 {
209         struct erase_priv_struct *priv = (void *)instr->priv;
210
211         if(instr->state != MTD_ERASE_DONE) {
212                 printk(KERN_WARNING "Erase at 0x%08x finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n", instr->addr, instr->state);
213                 jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
214         } else {
215                 jffs2_erase_succeeded(priv->c, priv->jeb);
216         }
217         kfree(instr);
218 }
219 #endif /* !__ECOS */
220
221 /* Hmmm. Maybe we should accept the extra space it takes and make
222    this a standard doubly-linked list? */
223 static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
224                         struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
225 {
226         struct jffs2_inode_cache *ic = NULL;
227         struct jffs2_raw_node_ref **prev;
228
229         prev = &ref->next_in_ino;
230
231         /* Walk the inode's list once, removing any nodes from this eraseblock */
232         while (1) {
233                 if (!(*prev)->next_in_ino) {
234                         /* We're looking at the jffs2_inode_cache, which is
235                            at the end of the linked list. Stash it and continue
236                            from the beginning of the list */
237                         ic = (struct jffs2_inode_cache *)(*prev);
238                         prev = &ic->nodes;
239                         continue;
240                 }
241
242                 if (SECTOR_ADDR((*prev)->flash_offset) == jeb->offset) {
243                         /* It's in the block we're erasing */
244                         struct jffs2_raw_node_ref *this;
245
246                         this = *prev;
247                         *prev = this->next_in_ino;
248                         this->next_in_ino = NULL;
249
250                         if (this == ref)
251                                 break;
252
253                         continue;
254                 }
255                 /* Not to be deleted. Skip */
256                 prev = &((*prev)->next_in_ino);
257         }
258
259         /* PARANOIA */
260         if (!ic) {
261                 JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
262                               " not found in remove_node_refs()!!\n");
263                 return;
264         }
265
266         D1(printk(KERN_DEBUG "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
267                   jeb->offset, jeb->offset + c->sector_size, ic->ino));
268
269         D2({
270                 int i=0;
271                 struct jffs2_raw_node_ref *this;
272                 printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n" KERN_DEBUG);
273
274                 this = ic->nodes;
275
276                 while(this) {
277                         printk( "0x%08x(%d)->", ref_offset(this), ref_flags(this));
278                         if (++i == 5) {
279                                 printk("\n" KERN_DEBUG);
280                                 i=0;
281                         }
282                         this = this->next_in_ino;
283                 }
284                 printk("\n");
285         });
286
287         switch (ic->class) {
288 #ifdef CONFIG_JFFS2_FS_XATTR
289                 case RAWNODE_CLASS_XATTR_DATUM:
290                         jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
291                         break;
292                 case RAWNODE_CLASS_XATTR_REF:
293                         jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
294                         break;
295 #endif
296                 default:
297                         if (ic->nodes == (void *)ic && ic->nlink == 0)
298                                 jffs2_del_ino_cache(c, ic);
299         }
300 }
301
302 void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
303 {
304         struct jffs2_raw_node_ref *block, *ref;
305         D1(printk(KERN_DEBUG "Freeing all node refs for eraseblock offset 0x%08x\n", jeb->offset));
306
307         block = ref = jeb->first_node;
308
309         while (ref) {
310                 if (ref->flash_offset == REF_LINK_NODE) {
311                         ref = ref->next_in_ino;
312                         jffs2_free_refblock(block);
313                         block = ref;
314                         continue;
315                 }
316                 if (ref->flash_offset != REF_EMPTY_NODE && ref->next_in_ino)
317                         jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
318                 /* else it was a non-inode node or already removed, so don't bother */
319
320                 ref++;
321         }
322         jeb->first_node = jeb->last_node = NULL;
323 }
324
325 static int jffs2_block_check_erase(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t *bad_offset)
326 {
327         void *ebuf;
328         uint32_t ofs;
329         size_t retlen;
330         int ret = -EIO;
331
332         if (c->mtd->point) {
333                 unsigned long *wordebuf;
334
335                 ret = c->mtd->point(c->mtd, jeb->offset, c->sector_size, &retlen, (unsigned char **)&ebuf);
336                 if (ret) {
337                         D1(printk(KERN_DEBUG "MTD point failed %d\n", ret));
338                         goto do_flash_read;
339                 }
340                 if (retlen < c->sector_size) {
341                         /* Don't muck about if it won't let us point to the whole erase sector */
342                         D1(printk(KERN_DEBUG "MTD point returned len too short: 0x%zx\n", retlen));
343                         c->mtd->unpoint(c->mtd, ebuf, jeb->offset, c->sector_size);
344                         goto do_flash_read;
345                 }
346                 wordebuf = ebuf-sizeof(*wordebuf);
347                 retlen /= sizeof(*wordebuf);
348                 do {
349                    if (*++wordebuf != ~0)
350                            break;
351                 } while(--retlen);
352                 c->mtd->unpoint(c->mtd, ebuf, jeb->offset, c->sector_size);
353                 if (retlen)
354                         printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08x\n",
355                                *wordebuf, jeb->offset + c->sector_size-retlen*sizeof(*wordebuf));
356                 return 0;
357         }
358  do_flash_read:
359         ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
360         if (!ebuf) {
361                 printk(KERN_WARNING "Failed to allocate page buffer for verifying erase at 0x%08x. Refiling\n", jeb->offset);
362                 return -EAGAIN;
363         }
364
365         D1(printk(KERN_DEBUG "Verifying erase at 0x%08x\n", jeb->offset));
366
367         for (ofs = jeb->offset; ofs < jeb->offset + c->sector_size; ) {
368                 uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
369                 int i;
370
371                 *bad_offset = ofs;
372
373                 ret = c->mtd->read(c->mtd, ofs, readlen, &retlen, ebuf);
374                 if (ret) {
375                         printk(KERN_WARNING "Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n", ofs, ret);
376                         goto fail;
377                 }
378                 if (retlen != readlen) {
379                         printk(KERN_WARNING "Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n", ofs, readlen, retlen);
380                         goto fail;
381                 }
382                 for (i=0; i<readlen; i += sizeof(unsigned long)) {
383                         /* It's OK. We know it's properly aligned */
384                         unsigned long *datum = ebuf + i;
385                         if (*datum + 1) {
386                                 *bad_offset += i;
387                                 printk(KERN_WARNING "Newly-erased block contained word 0x%lx at offset 0x%08x\n", *datum, *bad_offset);
388                                 goto fail;
389                         }
390                 }
391                 ofs += readlen;
392                 cond_resched();
393         }
394         ret = 0;
395 fail:
396         kfree(ebuf);
397         return ret;
398 }
399
400 static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
401 {
402         size_t retlen;
403         int ret;
404         uint32_t bad_offset;
405
406         switch (jffs2_block_check_erase(c, jeb, &bad_offset)) {
407         case -EAGAIN:   goto refile;
408         case -EIO:      goto filebad;
409         }
410
411         /* Write the erase complete marker */
412         D1(printk(KERN_DEBUG "Writing erased marker to block at 0x%08x\n", jeb->offset));
413         bad_offset = jeb->offset;
414
415         /* Cleanmarker in oob area or no cleanmarker at all ? */
416         if (jffs2_cleanmarker_oob(c) || c->cleanmarker_size == 0) {
417
418                 if (jffs2_cleanmarker_oob(c)) {
419                         if (jffs2_write_nand_cleanmarker(c, jeb))
420                                 goto filebad;
421                 }
422
423                 /* Everything else got zeroed before the erase */
424                 jeb->free_size = c->sector_size;
425         } else {
426
427                 struct kvec vecs[1];
428                 struct jffs2_unknown_node marker = {
429                         .magic =        cpu_to_je16(JFFS2_MAGIC_BITMASK),
430                         .nodetype =     cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
431                         .totlen =       cpu_to_je32(c->cleanmarker_size)
432                 };
433
434                 jffs2_prealloc_raw_node_refs(c, jeb, 1);
435
436                 marker.hdr_crc = cpu_to_je32(crc32(0, &marker, sizeof(struct jffs2_unknown_node)-4));
437
438                 vecs[0].iov_base = (unsigned char *) &marker;
439                 vecs[0].iov_len = sizeof(marker);
440                 ret = jffs2_flash_direct_writev(c, vecs, 1, jeb->offset, &retlen);
441
442                 if (ret || retlen != sizeof(marker)) {
443                         if (ret)
444                                 printk(KERN_WARNING "Write clean marker to block at 0x%08x failed: %d\n",
445                                        jeb->offset, ret);
446                         else
447                                 printk(KERN_WARNING "Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
448                                        jeb->offset, sizeof(marker), retlen);
449
450                         goto filebad;
451                 }
452
453                 /* Everything else got zeroed before the erase */
454                 jeb->free_size = c->sector_size;
455                 /* FIXME Special case for cleanmarker in empty block */
456                 jffs2_link_node_ref(c, jeb, jeb->offset | REF_NORMAL, c->cleanmarker_size, NULL);
457         }
458
459         down(&c->erase_free_sem);
460         spin_lock(&c->erase_completion_lock);
461         c->erasing_size -= c->sector_size;
462         c->free_size += jeb->free_size;
463         c->used_size += jeb->used_size;
464
465         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
466         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
467
468         list_add_tail(&jeb->list, &c->free_list);
469         c->nr_erasing_blocks--;
470         c->nr_free_blocks++;
471         spin_unlock(&c->erase_completion_lock);
472         up(&c->erase_free_sem);
473         wake_up(&c->erase_wait);
474         return;
475
476 filebad:
477         down(&c->erase_free_sem);
478         spin_lock(&c->erase_completion_lock);
479         /* Stick it on a list (any list) so erase_failed can take it
480            right off again.  Silly, but shouldn't happen often. */
481         list_add(&jeb->list, &c->erasing_list);
482         spin_unlock(&c->erase_completion_lock);
483         up(&c->erase_free_sem);
484         jffs2_erase_failed(c, jeb, bad_offset);
485         return;
486
487 refile:
488         /* Stick it back on the list from whence it came and come back later */
489         jffs2_erase_pending_trigger(c);
490         down(&c->erase_free_sem);
491         spin_lock(&c->erase_completion_lock);
492         list_add(&jeb->list, &c->erase_complete_list);
493         spin_unlock(&c->erase_completion_lock);
494         up(&c->erase_free_sem);
495         return;
496 }